Why

Why

Investigate the motivation and intent behind this repository's code. Why was it built this way? What alternatives were rejected?

This is not /research (official docs, specs, first-party APIs). This is not /gh-axi (GitHub operations).

Always available

Use these every run:

• git: blame, log --follow, merge commits
• repo docs: CONTEXT.md, docs/adr/, README, comments, tests
• gh if already authenticated: PR bodies, reviews, linked issues

Do not install Slack, Sentry, Datadog, Linear, Notion, or any other MCP from this skill. Optional sources only if the tool is already connected. A missing optional source is NOT_CONFIGURED, not a reason to add it.

Operating posture

• Evidence before narrative.
• Precision over polish. Cite or it is inference.
• Name the gaps.
• Do not infer intent from code shape alone.

Read references/epistemics.md for the confidence framework.

Step 1. Understand the target

Parse the target (file, pattern, feature, named decision) and the question (rationale, tradeoff, edge case, constraint, history). If vague, state your interpretation and proceed.

Step 2. Establish the code anchor

• File path(s) and line range(s)
• Key symbols
• Recent commits touching the target
• PR numbers from merge commits

git blame -L <start>,<end> <file>
git log --follow -p -- <file>
git log --oneline -20 -- <file>

If gh is authenticated:

gh pr view <number> --json title,body,author,createdAt,mergedAt,labels,closingIssuesReferences,comments,reviews

Step 3. Search what is actually available

Always: source control + repo docs.

If gh is authenticated: PRs and issues.

Optional, only if already connected: issue tracker MCP, long-form docs MCP, chat MCP, observability MCP, error tracker, analytics warehouse. Skip with an explicit NOT_CONFIGURED line. Do not spawn a panel of investigators for tools that are not there.

Step 4. Present

Keep this structure. Do not collapse confidence.

The Question. Restate what the user asked.

The Code in Question. File paths, line ranges, key symbols.

DIRECT EVIDENCE. Claims with citations (commit, PR, ADR, file:line).

INFERENCE. Claims supported by indirect evidence. Hedge. Explain the chain.

COMPETING HYPOTHESES. If the record fits more than one story, list them with evidence for and against.

UNKNOWN. Explicit gaps. Searches that returned empty.

SOURCES CONSULTED. One line per source, including the ones that returned nothing or were NOT_CONFIGURED.